"So, Tell Me About Your Policy...": Distillation of interpretable policies from Deep Reinforcement Learning agents

Giovanni Dispoto; Marcello Restelli; Paolo Bonetti

REVIEW 3 major objections 5 minor 1 cited by

"So, Tell Me About Your Policy...": Distillation of interpretable policies from Deep Reinforcement Learning agents

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper claims that a deep RL expert can be distilled into a linear policy by optimizing a lower bound on the value gap that emphasizes high-advantage decisions.

desk verdict The paper's core theoretical guarantee doesn't hold up (Eq. 6 squares a norm without justification), but the advantage-weighted distillation heuristic is a plausible idea with decent preliminary experiments. read the letter →

arxiv 2507.07848 v2 pith:I2HO6GNY submitted 2025-07-10 cs.LG

Giovanni Dispoto , Paolo Bonetti , Marcello Restelli This is my paper

classification cs.LG

keywords interpretablereinforcementlearningpolicydistillationbehavioralcloningadvantagefunctionlinearimitationfinancialtradingexplainability

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper proposes an algorithm, EXPLAIN, for turning a deep reinforcement-learning expert into a linear policy that a human can read, while keeping the expert's most consequential decisions. The central idea is to train the surrogate not by cloning every expert action, but by maximizing a lower bound on the value difference between surrogate and expert, a bound that weights imitation toward state-action pairs where the expert's advantage is large. The objective is derived from a known performance-difference inequality and is optimized entirely on a fixed dataset of expert trajectories, so no online access to the expert is needed. If the claim is right, then settings that cannot deploy black-box policies, such as regulated financial or safety-critical systems, could obtain transparent policies with a theoretical guarantee on how far their value can drift from the expert.

What carries the argument

The load-bearing object is the performance-difference lower bound of Eq. 6, an adaptation of a bound from [21]. In words, the return gap between expert and surrogate is at least the average advantage of the surrogate under the expert's state distribution minus a penalty that grows with the squared maximum distance between the two policies. Both terms become differentiable losses: the advantage term is optimized with a policy-gradient update weighted by the advantage (Lemma 5), and the distance term with a squared-difference behavioral-cloning loss (Lemma 6), combined in the EXPLAIN algorithm (Algorithm 1).

What would settle it

Evaluate the two sides of Eq. 6 on a fully-known small MDP for an expert and a surrogate whose policy distance is below one; any pair where the right-hand side exceeds the true difference $J^{\pi_I} - J^{\pi_E}$ would falsify the bound. A complementary check would be to train EXPLAIN on a task with rare high-advantage decisions and see whether the distilled policy still matches the expert on those states.

Watch

Extended reading notes

Core claim

On its own terms, the paper establishes that maximizing the lower bound $J^{\pi_I}_\mu - J^{\pi_E}_\mu \geq A^{\pi_I}_{\pi_E,\mu} - \frac{\gamma}{2(1-\gamma)^3} \lVert \Pi^{\pi_I} - \Pi^{\pi_E} \rVert_\infty^2$ yields an interpretable surrogate policy close to the expert, with the advantage term concentrating fidelity on high-impact decisions and the distance term preserving uniform similarity. The authors show that both terms have simple policy-gradient forms, so the surrogate can be trained on previously collected experience, and they demonstrate the recipe on classic control benchmarks and on a financial trading task where the resulting linear policy follows an intuitive volume-imbalance rule.

Load-bearing premise

The derivation assumes that replacing the policy-distance norm in the original bound with its square preserves the inequality; since the squared distance is smaller whenever the distance is below one, this is the step that, if invalid, removes the theoretical guarantee.

Editorial extensions

If this is right

A linear policy trained this way can be deployed in domains that prohibit black-box policies, since it is transparent and its training objective targets expert-level returns.
The method works with a fixed dataset collected once from the expert, so it does not require an online expert to label new states, unlike DAGGER-style approaches.
On the financial trading benchmark, the distilled policy reduces to a simple imbalance rule comparing first-level bid and ask sizes, giving a checkable summary of the deep agent.
In tasks with large discrete action spaces and compounding errors, the advantage-augmented objective empirically improves average return and reduces variance relative to behavioral cloning alone.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The same advantage-weighted objective likely transfers to other differentiable interpretable policy families, such as decision lists or generalized additive models, since nothing in the gradient derivation depends on the linear-softmax form.
The theoretical guarantee depends on accurate advantage estimates, so injecting controlled noise into the advantage values and measuring surrogate performance would reveal how gracefully the method degrades with critic error.
The trading results suggest the method could be used as a discovery tool for interpretable strategies, not only as a deployment artifact, since the extracted imbalance rule is a meaningful market signal on its own.
An alternative formulation that simply reweights the behavioral-cloning loss by the advantage might approximate the full objective with fewer hyperparameters; testing this would clarify whether the separate distance penalty is essential.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The short version: this is a well-motivated distillation method, but the theoretical bound that is supposed to support it is broken, and the algorithm as written doesn't match the lemmas. The idea is salvageable, but the paper needs major revision.

What's new: the specific combination of advantage-weighted policy gradient with a behavioral-cloning regularizer for offline distillation into a linear policy. This is a reasonable heuristic, and the experiments on discrete control and a trading environment do show that the method can beat plain BC, especially in the modified inverted pendulum. The paper is clearly written, and the code is available.

The main problem is Eq. 6. The paper goes from Corollary 4 to Eq. 6 by substituting ||q^πE||∞ ≤ 1/(1−γ). That substitution should put 1/(1−γ) in the numerator, and leave the norm to the first power. Instead, the norm gets squared with no explanation. Since policy distances lie in [0,1], squaring makes the penalty term smaller, so the claimed lower bound is actually larger than the valid one—it doesn't hold. The authors cite Pirotta et al. for the foundation, but that doesn't justify the algebraic move. So the 'theoretical guarantees' advertised in the abstract are not there. Additionally, the gradient in Lemma 6 has a factor 2, but Algorithm 1 drops it. And the expectations in Lemmas 5 and 6 are over actions sampled from πθ, while the algorithm samples from the expert dataset τ. That means the gradient estimates don't match the lemmas either. These are fixable in principle, but as written the math and the code are out of sync.

On the empirical side, the experiments are on toy control and a simplified trading task. That's fine for a workshop paper, but the comparison set is thin—mainly just BC with different η. There's no comparison to other interpretable distillation baselines like VIPER or PIRL, and no study of how the advantage estimates are obtained or how sensitive the result is to Q-value error. The trading environment lacks transaction costs, so the linear rule might not transfer to realistic settings. These are not fatal, but they limit the strength of the claims.

Who is this for: someone working on interpretable RL and policy distillation, especially those interested in advantage-weighting. The heuristic is worth knowing about, and the flawed derivation is a useful cautionary tale. I would not cite this as a theoretical contribution, but I might cite it for the empirical approach if I needed a baseline. A serious editor should send it to peer review—the issues are substantive and the authors need a chance to fix them—but under no circumstances should it be accepted as-is.

Referee Report

3 major / 5 minor

Summary. The paper proposes EXPLAIN, an offline algorithm that distills an interpretable linear policy from a deep RL expert by optimizing a weighted combination of an advantage term and a behavioral-cloning term. The authors claim this objective is supported by a theoretical lower bound on the expert-surrogate performance gap, adapted from Pirotta et al. [21]. The method is evaluated on two classic control tasks (discretized Inverted Pendulum and Mountain Car Continuous) and on a limit-order-book financial trading task, where the learned linear policy is shown to be interpretable and to outperform standard behavioral cloning in some settings. The paper includes open-source code and reports mean-variance results over six seeds.

Significance. If the theoretical guarantee were valid, the paper would contribute a useful principle for interpretable policy distillation: weighting imitation by the advantage function so that the surrogate focuses on state-action pairs that matter for return. The empirical case study on financial trading is compelling and the code availability is a plus. However, the central theoretical result is invalid as stated, and the algorithm's gradient computations do not implement the lemmas they cite. These flaws undermine the paper's main claimed contribution of providing theory-backed distillation. The empirical findings may still be of interest, but the manuscript needs substantial revision to either repair the theory or reframe the contributions as heuristic.

major comments (3)

[Section 4.1, Eq. (6)] The transition from Eq. (5) to Eq. (6) is mathematically unjustified. Substituting the bound ||q^{πE}||∞ ≤ 1/(1−γ) into Eq. (5) yields the penalty term −γ/(2(1−γ)^3) ||Π^{πI}−Π^{πE}||∞ with the norm to the first power, not squared. Since for any two policies ||Π^{πI}−Π^{πE}||∞ ∈ [0,1], we have ||x||²∞ ≤ ||x||∞, so −C||x||²∞ ≥ −C||x||∞; the right-hand side of Eq. (6) is therefore larger than the valid lower bound's right-hand side. A quantity larger than a valid lower bound need not itself be a lower bound, so Eq. (6) does not follow. The claim that maximizing Eq. (6) is supported by a performance guarantee is thus unsupported.
[Section 4.2, Algorithm 1] The gradient updates in Algorithm 1 do not match Lemmas 5 and 6. First, the ∇L(θ) line omits the factor 2 that appears in Lemma 6. Second, both lemmas compute expectations over a ∼ πθ (the surrogate policy), but the algorithm samples actions from the expert trajectories τ (i.e., a ∼ πE). Without importance-weight corrections, the computed gradients are biased estimators of the lemmas' gradients. Third, Definition 2 defines L(θ) as an unweighted sum over all states, while Lemma 6's proof replaces this with an expectation under d^{πE}; this is only valid if the state distribution is uniform, which is not the case in the experiments.
[Section 4.1, Definition 2] The paper justifies the replacement of the L∞ norm by the sum-of-squares loss L(θ) by noting that L(θ) is an upper bound of the squared L∞ norm. This is true but irrelevant: the valid bound in Eq. (5) contains the L∞ norm to the first power, and the sum of squared absolute differences is not an upper bound of that quantity. For example, a single state-action pair with absolute difference 0.5 gives L(θ) = 0.25 < 0.5, so minimizing L(θ) does not control the penalty term in the valid bound. This is a second, independent flaw in the theoretical derivation.

minor comments (5)

[Section 2] The transition model notation "P : S × A →P (S)" has a rendering issue and should be typeset as P : S × A → Δ(S).
[Algorithm 1] The gradient expressions in the algorithm use "Es∼τ" without specifying how the action a is sampled; please clarify that a is the expert's action in the collected tuple.
[Section 5.1, Figures 1 and 2] The captions state that the blue curve is "Adv+BC" and the orange is "BC", but the reader must infer that the y-axis is the mean episode return; please state the metric explicitly.
[Figure 4] The caption "Features Importance" should read "Feature Importance".
[Section 4.1] The display of Theorem 1 uses notation d^{πT_b}_μ, which is not defined; it would be clearer to write d^{π_b}_μ following the notation used later in Corollary 4.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central bound is adapted from an external published theorem, and the Eq. 6 squaring issue is a soundness gap, not a circular reduction.

full rationale

The paper's derivation chain is not circular. The central bound (Theorem 3, Corollary 4, Eq. 6) is an adaptation of Theorem 3.5 of Pirotta et al. [21], an externally published result; although M. Restelli is a co-author of [21], the bound is stated there with its own assumptions and is not defined in terms of the surrogate policy or the experimental outcomes, so citing it is real evidence rather than a self-referential premise. The algorithm's objective is a weighted combination of an advantage-weighted policy-gradient term (Lemma 5) and a behavioral-cloning distance term (Lemma 6); no parameter is fitted to the expert's return or to the reported performance, and eta is a free hyperparameter swept in the experiments. The financial-trading interpretation (Eq. 11) is read off the learned linear coefficients, not imposed as an input. The one substantive concern, that Eq. 6 squares the L-infinity policy-distance norm without justification so the claimed lower bound does not follow from Eq. 5, is a soundness gap rather than a circularity: the invalid step does not make the conclusion equal to an input by construction. Therefore no circular step is identified.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the prior bound of Pirotta et al. [21] (self-cited), on the availability of expert advantage estimates, and on the validity of Eq. 6, which is suspect. The main free parameter is η, tuned per environment. No new physical or conceptual entities are introduced.

free parameters (3)

η (BC penalization) = 0.01 for Inverted Pendulum, 1e-3 for Mountain Car
The weight of the behavioral cloning term is set separately per environment without a justification, and it controls the balance between advantage focus and uniform imitation. Selecting η by tuning on the target task turns the 'bound optimization' into a heuristic weighted sum.
λ (learning rate) = not specified
The initial learning rate for Adam is not given in the paper or the algorithm listing, so the optimization setup is incompletely specified.
N (number of iterations) = not specified
The number of optimization steps is left unspecified, which affects reproducibility.

assumptions (4)

standard math MDP formalism with discount factor γ < 1
All RL theory in the paper relies on the standard Markov Decision Process model and on the definitions of value functions, Q-functions, and advantages (Section 2).
domain assumption The expert's advantage function A^πE(s,a) is available and accurate
The algorithm requires advantage estimates for the expert policy. In practice, these are obtained from the expert's learned Q-function, which may be approximate, and the paper does not discuss the effect of estimation error.
standard math Theorem 1 from Pirotta et al. [21] is correct and applies unchanged
The paper's Theorem 3 and Corollary 4 are direct adaptations of this prior theorem. The authors do not reprove it, only state that the proof follows from [21].
standard math The sum-of-squares loss L(θ) is an upper bound of the squared L∞ norm
This is mathematically true because all terms are non-negative, but the use of this upper bound in Eq. 6 still depends on the invalid squaring of the original norm.

how reviews work

0 comments

Cite this review

Pith. "Pith review of "So, Tell Me About Your Policy...": Distillation of interpretable policies from Deep Reinforcement Learning agents." pith.science (2026). https://pith.science/paper/I2HO6GNY

@misc{pith2026250707848,
  author       = {Pith},
  title        = {Pith review of: "So, Tell Me About Your Policy...": Distillation of interpretable policies from Deep Reinforcement Learning agents},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/I2HO6GNY}},
  note         = {Machine review of arXiv:2507.07848}
}

read the original abstract

Recent advances in Reinforcement Learning (RL) largely benefit from the inclusion of Deep Neural Networks, boosting the number of novel approaches proposed in the field of Deep Reinforcement Learning (DRL). These techniques demonstrate the ability to tackle complex games such as Atari, Go, and other real-world applications, including financial trading. Nevertheless, a significant challenge emerges from the lack of interpretability, particularly when attempting to comprehend the underlying patterns learned, the relative importance of the state features, and how they are integrated to generate the policy's output. For this reason, in mission-critical and real-world settings, it is often preferred to deploy a simpler and more interpretable algorithm, although at the cost of performance. In this paper, we propose a novel algorithm, supported by theoretical guarantees, that can extract an interpretable policy (e.g., a linear policy) without disregarding the peculiarities of expert behavior. This result is obtained by considering the advantage function, which includes information about why an action is superior to the others. In contrast to previous works, our approach enables the training of an interpretable policy using previously collected experience. The proposed algorithm is empirically evaluated on classic control environments and on a financial trading scenario, demonstrating its ability to extract meaningful information from complex expert policies.

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

Selecting Feature Interactions for Generalized Additive Models by Distilling Foundation Models
cs.LG 2026-04 unverdicted novelty 7.0 of 10

TabDistill distills feature interactions from tabular foundation models via post-hoc attribution and inserts them into GAMs, yielding consistent predictive gains.

Reference graph

Works this paper leans on

34 extracted references · 14 canonical work pages · cited by 1 Pith paper

[21]

Pirotta, M

M. Pirotta, M. Restelli, A. Pecorino, and D. Calandriello. Safe policy iteration. In S. Dasgupta and D. McAllester, editors, Proceedings of the 30th International Conference on Machine Learning, volume 28 of Proceedings of Machine Learning Research , pages 307–315, Atlanta, Georgia, USA, 17–19 Jun 2013. PMLR. URL https://proceedings.mlr. press/v28/pirotta13.html

work page 2013
[1]

Exploratory Not Explanatory: Counterfactual Analysis of Saliency Maps for Deep Reinforcement Learning

A. Atrey, K. Clary, and D. D. Jensen. Exploratory not explana- tory: Counterfactual analysis of saliency maps for deep RL. CoRR, abs/1912.05743, 2019. URL http://arxiv.org/abs/1912.05743

work page Pith review arXiv 1912
[2]

L. J. Ba and R. Caruana. Do deep nets really need to be deep? In Z. Ghahramani, M. Welling, C. Cortes, N. Lawrence, and K. Weinberger, editors, Advances in Neural Information Pro- cessing Systems , volume 27. Curran Associates, Inc., 2014. URL https://proceedings.neurips.cc/paper_files/paper/2014/file/ b0c355a9dedccb50e5537e8f2e3f0810-Paper.pdf

work page 2014
[3]

Bastani, Y

O. Bastani, Y . Pu, and A. Solar-Lezama. Verifiable reinforcement learn- ing via policy extraction. In S. Bengio, H. Wallach, H. Larochelle, K. Grauman, N. Cesa-Bianchi, and R. Garnett, editors, Advances in Neural Information Processing Systems, volume 31. Curran Associates, Inc., 2018. URL https://proceedings.neurips.cc/paper_files/paper/2018/ file/e6d85...

work page 2018
[4]

L. Breiman. Random forests. Mach. Learn., 45(1):5–32, Oct. 2001. ISSN 0885-6125. doi: 10.1023/A:1010933404324. URL https://doi. org/10.1023/A:1010933404324

doi:10.1023/a:1010933404324 2001
[6]

Chen and C

T. Chen and C. Guestrin. Xgboost: A scalable tree boosting system. CoRR, abs/1603.02754, 2016. URL http://arxiv.org/abs/1603.02754

arXiv 2016
[7]

F. Cruz, R. Dazeley, and P. Vamplew. Memory-based explainable rein- forcement learning. In J. Liu and J. Bailey, editors, AI 2019: Advances in Artificial Intelligence , pages 66–77, Cham, 2019. Springer Interna- tional Publishing. ISBN 978-3-030-35288-2

work page 2019
[8]

Das and P

A. Das and P. Rad. Opportunities and challenges in explainable artificial intelligence (XAI): A survey.CoRR, abs/2006.11371, 2020. URL https: //arxiv.org/abs/2006.11371

arXiv 2006

Show all 34 references

[9]

Ernst, P

D. Ernst, P. Geurts, and L. Wehenkel. Tree-based batch mode reinforce- ment learning. Journal of Machine Learning Research, 6(18):503–556,
[10]

M. D. Gould, M. A. Porter, S. Williams, M. McDonald, D. J. Fenn, and S. D. Howison. Limit order books, 2013. URL https://arxiv.org/abs/ 1012.0349

2013 arXiv
[11]

Haarnoja, A

T. Haarnoja, A. Zhou, P. Abbeel, and S. Levine. Soft actor-critic: Off- policy maximum entropy deep reinforcement learning with a stochastic actor. CoRR, abs/1801.01290, 2018. URL http://arxiv.org/abs/1801. 01290

2018 arXiv
[12]

Ho and S

J. Ho and S. Ermon. Generative adversarial imitation learning. CoRR, abs/1606.03476, 2016. URL http://arxiv.org/abs/1606.03476

2016 arXiv
[13]

Kaplan, S

J. Kaplan, S. McCandlish, T. Henighan, T. B. Brown, B. Chess, R. Child, S. Gray, A. Radford, J. Wu, and D. Amodei. Scaling laws for neural language models. CoRR, abs/2001.08361, 2020. URL https://arxiv.org/abs/2001.08361

2001 arXiv
[14]

D. P. Kingma and J. Ba. Adam: A method for stochastic optimiza- tion. CoRR, abs/1412.6980, 2014. URL https://api.semanticscholar. org/CorpusID:6628106

2014 arXiv
[15]

Likmeta, A

A. Likmeta, A. M. Metelli, A. Tirinzoni, R. Giol, M. Restelli, and D. Romano. Combining reinforcement learning with rule- based controllers for transparent and general decision-making in autonomous driving. Robotics and Autonomous Systems , 131: 103568, 2020. ISSN 0921-8890. d...

2020
[17]

S. M. Lundberg and S. Lee. A unified approach to interpreting model predictions. CoRR, abs/1705.07874, 2017. URL http://arxiv.org/abs/ 1705.07874

2017 arXiv
[18]

Madumal, T

P. Madumal, T. Miller, L. Sonenberg, and F. Vetere. Explainable re- inforcement learning through a causal lens. In The Thirty-Fourth AAAI Conference on Artificial Intelligence, AAAI 2020, The Thirty-Second In- novative Applications of Artificial Intelligence Conference, IAAI 2...

2020
[19]

Milani, N

S. Milani, N. Topin, M. Veloso, and F. Fang. Explainable reinforcement learning: A survey and comparative review. ACM Comput. Surv., 56 (7), Apr. 2024. ISSN 0360-0300. doi: 10.1145/3616864. URL https: //doi.org/10.1145/3616864

2024 doi
[20]

V . Mnih, K. Kavukcuoglu, D. Silver, A. Graves, I. Antonoglou, D. Wier- (a) Short (b) Long Figure 6: Weights evolution of the softmax regression related to each feature. In Figure 6a are visible the weights related to the short action, and in Figure 6b, the weights of the long...

2013 arXiv
[22]

Raffin, A

A. Raffin, A. Hill, A. Gleave, A. Kanervisto, M. Ernestus, and N. Dor- mann. Stable-baselines3: Reliable reinforcement learning implementa- tions. Journal of Machine Learning Research, 22(268):1–8, 2021. URL http://jmlr.org/papers/v22/20-1364.html

2021
[23]

Ross and D

S. Ross and D. Bagnell. Efficient reductions for imitation learning. In Y . W. Teh and M. Titterington, editors, Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics , vol- ume 9 of Proceedings of Machine Learning Research, pages 661...

2010
[24]

S. Ross, G. J. Gordon, and J. A. Bagnell. No-regret reductions for imi- tation learning and structured prediction. CoRR, abs/1011.0686, 2010. URL http://arxiv.org/abs/1011.0686

2010 arXiv
[25]

A. A. Rusu, S. G. Colmenarejo, Ç. Gülçehre, G. Desjardins, J. Kirk- patrick, R. Pascanu, V . Mnih, K. Kavukcuoglu, and R. Hadsell. Pol- icy distillation. In Y . Bengio and Y . LeCun, editors, 4th Interna- tional Conference on Learning Representations, ICLR 2016, San Juan, Puer...

2016 arXiv
[26]

A. E. Sallab, M. Abdou, E. Perot, and S. Yogamani. Deep reinforcement learning framework for autonomous driving. Electronic Imaging, 29 (19):70–76, Jan. 2017. ISSN 2470-1173. doi: 10.2352/issn.2470-1173. 2017.19.avm-023. URL http://dx.doi.org/10.2352/ISSN.2470-1173. 2017.19.A VM-023

2017 doi
[27]

Silver, A

D. Silver, A. Huang, C. J. Maddison, A. Guez, L. Sifre, G. van den Driessche, J. Schrittwieser, I. Antonoglou, V . Panneershelvam, M. Lanctot, S. Dieleman, D. Grewe, J. Nham, N. Kalchbrenner, I. Sutskever, T. Lillicrap, M. Leach, K. Kavukcuoglu, T. Graepel, and D. Hassabis. Ma...

2016
[28]

R. S. Sutton and A. G. Barto. Reinforcement learning: An introduction. MIT press, 2018

2018
[29]

R. S. Sutton, D. McAllester, S. Singh, and Y . Mansour. Policy gradi- ent methods for reinforcement learning with function approximation. In Proceedings of the 13th International Conference on Neural Informa- tion Processing Systems, NIPS’99, page 1057–1063, Cambridge, MA, USA...

1999
[30]

Towers, A

M. Towers, A. Kwiatkowski, J. Terry, J. U. Balis, G. De Cola, T. Deleu, M. Goulão, A. Kallinteris, M. Krimmel, A. KG, et al. Gymnasium: A standard interface for reinforcement learning environments. arXiv preprint arXiv:2407.17032, 2024

2024 arXiv
[32]

R. J. Williams. Simple statistical gradient-following algorithms for connectionist reinforcement learning. Mach. Learn., 8(3–4):229–256, May 1992. ISSN 0885-6125. doi: 10.1007/BF00992696. URL https: //doi.org/10.1007/BF00992696

1992 doi
[33]

J. Xing, T. Nagata, X. Zou, E. Neftci, and J. L. Krichmar. Achieving efficient interpretability of reinforcement learning via policy distilla- tion and selective input gradient regularization. Neural Networks, 161: 228–241, 2023. ISSN 0893-6080. doi: https://doi.org/10.1016/j....

2023 doi
[35]

URL http://arxiv.org/abs/1804.02477

arXiv
[2005]

URL http://jmlr.org/papers/v6/ernst05a.html
[2018]

URL http://arxiv.org/abs/1807.05887

arXiv
[2021]

URL https://arxiv.org/abs/2101.07107

arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [21]

Pirotta, M

M. Pirotta, M. Restelli, A. Pecorino, and D. Calandriello. Safe policy iteration. In S. Dasgupta and D. McAllester, editors, Proceedings of the 30th International Conference on Machine Learning, volume 28 of Proceedings of Machine Learning Research , pages 307–315, Atlanta, Georgia, USA, 17–19 Jun 2013. PMLR. URL https://proceedings.mlr. press/v28/pirotta13.html

work page 2013

[2] [1]

Exploratory Not Explanatory: Counterfactual Analysis of Saliency Maps for Deep Reinforcement Learning

A. Atrey, K. Clary, and D. D. Jensen. Exploratory not explana- tory: Counterfactual analysis of saliency maps for deep RL. CoRR, abs/1912.05743, 2019. URL http://arxiv.org/abs/1912.05743

work page Pith review arXiv 1912

[3] [2]

L. J. Ba and R. Caruana. Do deep nets really need to be deep? In Z. Ghahramani, M. Welling, C. Cortes, N. Lawrence, and K. Weinberger, editors, Advances in Neural Information Pro- cessing Systems , volume 27. Curran Associates, Inc., 2014. URL https://proceedings.neurips.cc/paper_files/paper/2014/file/ b0c355a9dedccb50e5537e8f2e3f0810-Paper.pdf

work page 2014

[4] [3]

Bastani, Y

O. Bastani, Y . Pu, and A. Solar-Lezama. Verifiable reinforcement learn- ing via policy extraction. In S. Bengio, H. Wallach, H. Larochelle, K. Grauman, N. Cesa-Bianchi, and R. Garnett, editors, Advances in Neural Information Processing Systems, volume 31. Curran Associates, Inc., 2018. URL https://proceedings.neurips.cc/paper_files/paper/2018/ file/e6d85...

work page 2018

[5] [4]

L. Breiman. Random forests. Mach. Learn., 45(1):5–32, Oct. 2001. ISSN 0885-6125. doi: 10.1023/A:1010933404324. URL https://doi. org/10.1023/A:1010933404324

doi:10.1023/a:1010933404324 2001

[6] [6]

Chen and C

T. Chen and C. Guestrin. Xgboost: A scalable tree boosting system. CoRR, abs/1603.02754, 2016. URL http://arxiv.org/abs/1603.02754

arXiv 2016

[7] [7]

F. Cruz, R. Dazeley, and P. Vamplew. Memory-based explainable rein- forcement learning. In J. Liu and J. Bailey, editors, AI 2019: Advances in Artificial Intelligence , pages 66–77, Cham, 2019. Springer Interna- tional Publishing. ISBN 978-3-030-35288-2

work page 2019

[8] [8]

Das and P

A. Das and P. Rad. Opportunities and challenges in explainable artificial intelligence (XAI): A survey.CoRR, abs/2006.11371, 2020. URL https: //arxiv.org/abs/2006.11371

arXiv 2006